• SDSN - Greece
• Publications close
• Hydrology and Earth System Sciences... close
• Hydrology and Earth System Sciences close

Despite high potential benefits, the development of seasonal forecasting tools in the water sector has been slower than in other sectors. Here we assess the skill of seasonal forecasting tools for lakes and reservoirs set up at four sites in Australia and Europe. These tools consist of coupled hydrological catchment and lake models forced with seasonal meteorological forecast ensembles to provide probabilistic predictions of seasonal anomalies in water discharge, temperature and ice-off. Successful implementation requires a vigorous assessment of the tools? predictive skill and an apportionment of the predictability between legacy effects and input forcing data. To this end, models were forced with two meteorological datasets from the European Centre for Medium-Range Weather Forecasts (ECMWF), the seasonal forecasting system, SEAS5, with 3-month lead times and the ERA5 reanalysis. Historical skill was assessed by comparing both model outputs, i.e. seasonal lake hindcasts (forced with SEAS5), and pseudo-observations (forced with ERA5). The skill of the seasonal lake hindcasts was generally low although higher than the reference hindcasts, i.e. pseudo-observations, at some sites for certain combinations of season and variable. The SEAS5 meteorological predictions showed less skill than the lake hindcasts. In fact, skilful lake hindcasts identified for selected seasons and variables were not always synchronous with skilful SEAS5 meteorological hindcasts, raising questions on the source of the predictability.A set of sensitivity analyses showed that most of the forecasting skill originates from legacy effects, although during winter and spring in Norway some skill was coming from SEAS5 over the 3-month target season. When SEAS5 hindcasts were skilful, additional predictive skill originates from the interaction between legacy and SEAS5 skill. We conclude that lake forecasts forced with an ensemble of boundary conditions resampled from historical meteorology are currently likely to yield higher-quality forecasts in most cases. This research has been supported by the Norges Forskningsråd (grant no. 274208), the European Union (grant no. 690462) and MINECO-AEI (projects PCIN-2017-062 and PCIN-2017-092).

Abstract. Despite high potential benefits, the development of seasonal forecasting tools in the water sector has been slower than in other sectors. Here we assess the skill of seasonal forecasting tools for lake and reservoir set up at four sites in Australia and Europe. These tools, as previously presented, consist of coupled hydrological catchment and lake models forced with seasonal climate forecast ensembles to provide probabilistic predictions of seasonal anomalies in water discharge, temperature and ice-off. Successful implementation requires a rigorous assessment of the tools’ predictive skill and an apportionment of the predictability between legacy effects and input data. To this end, models were forced with two meteorological datasets from the European Centre for Medium Range Weather Forecasts (ECMWF), the seasonal forecasts SEAS5 and the ERA5 reanalysis. Historical skill was assessed by comparing both model outputs, i.e., seasonal lake hindcasts (forced with SEAS5) and pseudo-observations (forced with ERA5). The skill of the seasonal lake hindcasts was generally low, but higher than SEAS5 climate hindcasts. Nevertheless, lake and SEAS5 windows of opportunity were identified, although they were not always synchronous, raising questions on the source of the predictability. A set of sensitivity analyses showed that most of the forecasting skill originates from legacy effects, although during winter and spring in Norway some skill was coming from SEAS5 over the target season. When SEAS5 hindcasts were skillful, additional predictability originates from the interaction between legacy and SEAS5 skill. We conclude that a climatology driven forecast is currently likely to yield higher quality forecasts.

Freshwater management is challenging, and advance warning that poor water quality was likely, a season ahead, could allow for preventative measures to be put in place. To this end, we developed a Bayesian network (BN) for seasonal lake water quality prediction. BNs have become popular in recent years, but the vast majority are discrete. Here, we developed a Gaussian Bayesian network (GBN), a simple class of continuous BN. The aim was to forecast, in spring, mean total phosphorus (TP) and chlorophyll a (chl a) concentration, mean water colour, and maximum cyanobacteria biovolume for the upcoming growing season (May–October) in Vansjø, a shallow nutrient-rich lake in southeastern Norway. To develop the model, we first identified controls on interannual variability in seasonally aggregated water quality. These variables were then included in a GBN, and conditional probability densities were fit using observations (≤39 years). GBN predictions had R2 values of 0.37 (chl a) to 0.75 (colour) and classification errors of 32 % (TP) to 17 % (cyanobacteria). For all but lake colour, including weather variables did not improve the predictive performance (assessed through cross-validation). Overall, we found the GBN approach to be well suited to seasonal water quality forecasting. It was straightforward to produce probabilistic predictions, including the probability of exceeding management-relevant thresholds. The GBN could be sensibly parameterised using only the observed data, despite the small dataset. Developing a comparable discrete BN was much more subjective and time-consuming. Although low interannual variability and high temporal autocorrelation in the study lake meant the GBN performed only slightly better than a seasonal naïve forecast (where the forecasted value is simply the value observed the previous growing season), we believe that the forecasting approach presented here could be particularly useful in areas with higher sensitivity to catchment nutrient delivery and seasonal climate and for forecasting at shorter (daily or monthly) timescales. Despite the parametric constraints of GBNs, their simplicity, together with the relative accessibility of BN software with GBN handling, means they are a good first choice for BN development with continuous variables.

A climatic anomaly can potentially affect the hydrological behaviour of a catchment for several years. This article presents a new approach to quantifying this multi-year hydrological memory, using exclusively streamflow and climate data. Rather than providing a single value of catchment memory, we aim to describe how this memory fades over time. The precipitation-runoff relationship is analyzed through the concept of elasticity. Elasticity quantifies the change in one quantity caused by the change in another quantity. We analyze the elasticity of the relation between the annual anomalies of runoff yield and humidity index. We identify Catchment Forgetting Curves (CFC) to quantify multi-year catchment memory, considering not only the current year's humidity anomaly but also the anomalies of the preceding years. The variability of CFCs is investigated on a set of 158 Swedish and 527 French catchments. As expected, French catchments overlying large aquifers exhibit a long memory, i.e., with the impact of climate anomalies detected over several years. In Sweden, the expected effect of the lakes is less clear. For both countries, a relatively strong relationship between the humidity index and memory is identified, with drier regions exhibiting longer memory. Taking into account the multi-year memory has significantly improved the elasticity analysis for 15 % of the catchments. Our work thus underlines the need to account for catchment memory in order to produce meaningful and geographically coherent elasticity indices. International audience

Abstract. Climate warming in mountain regions is resulting in glacier shrinking, seasonal snow cover reduction, and changes in the amount and seasonality of meltwater runoff, with consequences on water availability. Droughts are expected to become more severe in the future with economical and environmental losses both locally and downstream. Effective adaptation strategies involve multiple timescales, and seasonal forecasts can help in the optimization of the available snow and water resources with a lead time of several months. We developed a prototype to generate seasonal forecasts of snow depth and snow water equivalent with a starting date of 1 November and a lead time of 7 months, so up to 31 May of the following year. The prototype has been co-designed with end users in the field of water management, hydropower production and mountain ski tourism, meeting their needs in terms of indicators, time resolution of the forecasts and visualization of the forecast outputs. In this paper we present the modelling chain, based on the seasonal forecasts of the ECMWF and Météo-France seasonal prediction systems, made available through the Copernicus Climate Change Service (C3S) Climate Data Store. Seasonal forecasts of precipitation, near-surface air temperature, radiative fluxes, wind and relative humidity are bias-corrected and downscaled to three sites in the Western Italian Alps and finally used as input for the physically based multi-layer snow model SNOWPACK. Precipitation is bias-corrected with a quantile mapping method using ERA5 reanalysis as a reference and then downscaled with the RainFARM stochastic procedure in order to allow an estimate of uncertainties due to the downscaling method. The impacts of precipitation bias adjustment and downscaling on the forecast skill have been investigated. The skill of the prototype in predicting the deviation of monthly snow depth with respect to the normal conditions from November to May in each season of the hindcast period 1995–2015 is demonstrated using both deterministic and probabilistic metrics. Forecast skills are determined with respect to a simple forecasting method based on the climatology, and station measurements are used as reference data. The prototype shows good skills at predicting the tercile category, i.e. snow depth below and above normal, in the winter (lead times: 2–3–4 months) and spring (lead times: 5–6–7 months) ahead: snow depth is predicted with higher accuracy (Brier skill score) and higher discrimination (area under the relative operating characteristics (ROC) curve skill score) with respect to a simple forecasting method based on the climatology. Ensemble mean monthly snow depth forecasts are significantly correlated with observations not only at short lead times of 1 and 2 months (November and December) but also at lead times of 5 and 6 months (March and April) when employing the ECMWFS5 forcing. Moreover the prototype shows skill at predicting extremely dry seasons, i.e. seasons with snow depth below the 10th percentile, while skills at predicting snow depth above the 90th percentile are model-, station- and score-dependent. The bias correction of precipitation forecasts is essential in the case of large biases in the global seasonal forecast system (MFS6) to reconstruct a realistic snow depth climatology; however, no remarkable differences are found among the skill scores when the precipitation input is bias-corrected, downscaled, or bias-corrected and downscaled, compared to the case in which raw data are employed, suggesting that skill scores are weakly sensitive to the treatment of the precipitation input.

Groundwater levels (GWLs) very often fluctuate over a wide range of timescales (intra-annual, annual, multi-annual, and decadal). In many instances, aquifers act as low-pass filters, dampening the high-frequency variability and amplifying low-frequency variations (from multi-annual to decadal timescales) which basically originate from large-scale climate variability. Within the aim of better understanding and ultimately anticipating groundwater droughts and floods, it appears crucial to evaluate whether (and how much) the very high or very low GWLs are resulting from such low-frequency variability (LFV), which was the main objective of the study presented here. As an example, we focused on exceedance and non-exceedance of the 80 % and 20 % GWL percentiles respectively, in the Paris Basin aquifers over the 1976–2019 period. GWL time series were extracted from a database consisting of relatively undisturbed GWL time series regarding anthropogenic influence (water abstraction by either continuous or periodic pumping) over metropolitan France. Based on this dataset, our approach consisted in exploring the effect of GWL low-frequency components on threshold exceedance and non-exceedance by successively filtering out low-frequency components of GWL signals using maximum overlap discrete wavelet transform (MODWT). Multi-annual (∼7-year) and decadal (∼17-year) variabilities were found to be the predominant LFVs in GWL signals, in accordance with previous studies in the northern France area. By filtering out these components (either independently or jointly), it is possible to (i) examine the proportion of high-level (HL) and low-level (LL) occurrences generated by these variabilities and (ii) estimate the contribution of each of these variabilities in explaining the occurrence of major historical events associated with well-recognized societal impacts. A typology of GWL variations in Paris Basin aquifers was first determined by quantifying the variance distribution across timescales. Four GWL variation types could be found according to the predominance of annual, multi-annual, or/and decadal variabilities in these signals: decadal-dominant (type iD), multi-annual- and decadal-dominant (type iMD), annual-dominant (type cA), and annual- and multi-annual-dominant (type cAM). We observed a clear dependence of high and low GWL on LFV for aquifers exhibiting these four GWL variation types. In addition, the respective contribution of multi-annual and decadal variabilities in the threshold exceedance varied according to the event. In numerous aquifers, it also appeared that the sensitivity to LFV was higher for LLs than HLs. A similar analysis was conducted on the only available long-term GWL time series which covered a hundred years. This allowed us to highlight the potential influence of multidecadal variability on HLs and LLs too. This study underlined the key role of LFV in the occurrence of HLs and LLs. Since LFV originates from large-scale stochastic climate variability as demonstrated in many previous studies in the Paris Basin or nearby regions, our results point out that (i) poor representation of LFV in general circulation model (GCM) outputs used afterwards for developing hydrological projections can result in strong uncertainty in the assessment of future groundwater extremes (GWEs), and (ii) potential changes in the amplitude of LFV, be they natural or induced by global climate change, may lead to substantial changes in the occurrence and severity of GWEs for the next decades. Finally, this study also stresses the fact that due to the stochastic nature of LFV, no deterministic prediction of future GWEs for the mid- or long-term horizons can be achieved, even though LFV may look periodic.

Abstract. Freshwater management is challenging, and advance warning that poor water quality was likely, a season ahead, could allow for preventative measures to be put in place. To this end, we developed a Bayesian network (BN) for seasonal lake water quality prediction. BNs have become popular in recent years, but the vast majority are discrete. Here we developed a Gaussian Bayesian network (GBN), a simple class of continuous BN. The aim was to forecast, in spring, total phosphorus (TP), chlorophyll-a (chl-a), cyanobacteria biovolume and water colour for the coming growing season (May–October) in lake Vansjø in southeast Norway. To develop the model, we first identified controls on inter-annual variability in water quality using correlations, scatterplots, regression tree based feature importance analysis and process knowledge. Key predictors identified were lake conditions the previous summer, a TP control on algal variables, a colour-cyanobacteria relationship, and weaker relationships between precipitation and colour and between wind and chl-a. These variables were then included in the GBN and conditional probability densities were fitted using observations (≤ 39 years). GBN predictions had R2 values of 0.37 (cyanobacteria) to 0.75 (colour) and classification errors of 32 % (TP) to 13 % (cyanobacteria). For all but lake colour, including weather nodes did not improve predictive performance (assessed through cross validation). Overall, we found the GBN approach to be well-suited to seasonal water quality forecasting. It was straightforward to produce probabilistic predictions, including the probability of exceeding management-relevant thresholds. The GBN could be purely parameterised using observed data, despite the small dataset. This wasn’t possible using a discrete BN, highlighting a particular advantage of using GBNs when sample sizes are small. Although low interannual variability and high temporal autocorrelation in the study lake meant the GBN performed similarly to a seasonal naïve forecast, we believe the forecasting approach presented could be useful in areas with higher sensitivity to catchment nutrient delivery and seasonal climate, and for forecasting at shorter time scales (e.g. daily to monthly). Despite the parametric constraints of GBNs, their simplicity, together with the relative accessibility of BN software with GBN handling, means they are a good first choice for BN development, particularly when datasets for model training are small.